[infinispan-dev] NBST design

[Mircea Markus]

Another aspect Erik raised is the TACH/server hinting (rack, machine in particular) - I guess these would be parameters of the CH factory? 
Also a notes on the RoutingTable: it is a more effective data structure than the old CH, as it basically keeps a cache with all the possible outcomes of the CH function (instead of building the collection of owner on every call)

On 15 Aug 2012, at 12:35, Adrian Nistor wrote:
> Hi Galder,
> 
> My (superficial) understanding of the HR v1.0 and v1.1 is that clients 
> will actually derive the CH of the cluster based on the received member 
> address list + virtual node info.
> Now with in the new CH algo we do not use the address of members in the 
> computation and the concept of virtual nodes has vanished.
> So my feeling is we need a HR v1.2 where the full CH (the routing table) 
> si sent to the client as you said. Are you worried about sending the 
> routing table for reasons other than backward compatibility?
> 
> Regarding backward compat, I know Dan Berindei has done some wizardry 
> there to be able to support old clients, but I do not know any details. 
> He will hopefully clear things up, and it seems we also need to update 
> the HR protocol wiki page regarding this aspect.